探索網路新聞主題與政黨傾向：語意網絡分析之環境公投研究應用 - 政大學術集成

全文

(1)國立政治大學國際傳播英語碩士學位學程 International Master’s Program in International Communication Studies College of Communication National Chengchi University. 碩士論文政治大 Master’s Thesis. 立. ‧ 國. 學. ‧. 探索網路新聞主題與政黨傾向：語意網絡分析之環境公投研究應用. y. Nat. sit. n. al. er. io. Explore Partisan Slant of Online News Framing: A Semantic Network Analysis on an Environmental Referendum Issue. Ch. engchi. i Un. v. Student: 高千容 (Chien Jung, Kao) Advisor: 林翠絹 (Trisha Tsui-Chuan Lin, Ph.D.). 中華民國 109 年 2 月 February, 2020. DOI:10.6814/NCCU202000381.

(2) 探索網路新聞主題與政黨傾向：語意網絡分析之環境公投研究應用 Explore Partisan Slant of Online News Framing: A Semantic Network Analysis on an Environmental Referendum Issue. 研究生：高千容指導教授：林翠絹. Advisor: Trisha Tsui-Chuan Lin, Ph.D.. 政治大國立政治大學. 國際傳播英語碩士學位學程. 學. 碩士論文. ‧. ‧ 國. 立. Student: Chien Jung, Kao. sit. y. Nat. n. er. io. A Thesis a lInternational Master’s Program Submitted to in v i n C hCommunicationUStudies International engchi National Chengchi University In Partial Fulfillment of the Requirement For the degree of Master of Arts. 中華民國 109 年 2 月 February, 2020. DOI:10.6814/NCCU202000381.

(3) ACKNOWLEDGEMENT. I would like to express my deepest gratitude to my supervisor, Dr. Trisha T. C. Lin, for her expertise, encouragement and support. Without her continuous and generous guidance, this thesis would not be possible to complete. I also extend my appreciation to the committee members, Dr. Wen Cheng, Fu and Dr. Yu Chung, Cheng for providing constructive and kind suggestions and feedback, which made the thesis more completed and insightful.. Special appreciations are. 治政 for QSearch and Taiwan 大 Institute 立. for Governance and. ‧ 國. 學. Communication Research (TIGCR) for kindly and generously providing Facebook dataset and news media dataset. Without the two datasets provided by QSearch and TIGCR, this study. ‧. would not proceed smoothly and efficiently in this way. In addition, I would like to thank to. y. Nat. n. al. er. io. sit. my supervisor again, for authorizing the access of the news media dataset provided by TIGCR.. Ch. i Un. v. I would like to especially express thanks to Shih En, Chou, the Chief Technology Officer (CTO). engchi. of QSearch (Dolyman Inc.), for the invaluable suggestions to this research. Last but not least, I express my sincere acknowledgement to my family, friends, intern colleagues, classmates, school administrators, and anyone else who gave me support and company during the journey of my graduate studies.. DOI:10.6814/NCCU202000381.

(4) Abstract In Taiwan, the project of “Shenao thermal power plant” expansion (Ministry of Economics Affairs, 2018) was opposed by environmental groups and local residents (Liao, 2018; Greenpeace, 2018; DeAeth, 2018). The divergent discussions on Shenao power plant issue was voted in in 2018 Referendum in order to seek public opinions (Ferry, 2018b). As politics in Taiwan is characterized as a two-party system consisting of Kuo Ming Tang (KMT)led political alliance and Democracy Progress Party-led (DPP) league (Hsiao, 2014), perspectives about environmental issues inevitably have been divided into two types of claims supported by different political parties and ideologies. The news reporting about Shenao power plant in both traditional and digital-native media showed struggles and conflicts between KMT and DPP (ETtoday News, 2018; Hsu, 2018; Chang, 2018). Slant or bias in news involves selections of news media contents (Schudson, 2003). Some mainstream media showed partisan slant which were in favor of a political party (Merkley, 2019). Past studies found it difficult to operationalize concepts of “bias” or “balance” (Hopmann, Van Aelst & Legnante, 2012; Lichter, 2017). Researchers focused on content-based analysis, considering that framing or text-based factors as a crucial factor of slant. In order to cover large amount of news contents, these studies only measure partisan slant based on frequency of mentions. Alternatively, this research proposed to use semantic network analysis to examine the selected issues regarding partisan slant in news media framing. Before exploring the partisan slant, this research investigated news media’s framing in environmental news coverage during 2018 referendum. This study’s RQ1 examined the salient issues and framing of the Shenao power plant issues in different media types (traditional news media, digital-native news media, and social media). It further compared salient issues of Facebook posts from fan pages of KMT/DPP political actors (politicians or political organizations) and environmental group fan pages related to the Shenao power plant issue. RQ2 explored the partisan slant of news media on Shenao power plant referendum issue, which regarded social media contents from two political parties (KMT and DPP) as the parameters to measure news media’s partisan slant. The environmental groups were used as the non-partisan parameter for comparison. RQ2a measured the partisan slant of two types of news media, while RQ2b identified partisan slant of all news media. The semantic network correlation analysis utilized Quadratic Assignment Procedure (QAP) (Krackhardt, 1988) to identify the partisan leaning of news media. Finally, RQ3 investigated the salient issues in news media which were labelled their partisan leaning as pro-KMT or pro-DPP categories. This research conducted analysis with two datasets which included uMiner’s online media news articles and QSearch’s Facebook posts dated July 1 to November 30, 2018. This research considered the Environmental Protect Administration’s (EPA) official announcement on June 28 that Shenao power plant expansion project was approved by environmental impact assessment review as the the start for data collection, and it ended at six days after the referendum on 24 November 2018. Through Boolean keywords, uMiner platform provided news contents regarding Shenao power plant issue, consisting of 2,627 articles, including 1,597 articles from six traditional media brands (61%) and 1,030 articles from four digital native news media (39%). QSearch Trend panel provided 817 Facebook posts related to Shenao power plant (550 from KMT, 45 posts from DPP, and 222 from environmental group fan pages). As for data analysis, RQ1 used the two dataset, while RQ2, and RQ3 used the same news media dataset but different Facebook dataset. In order eliminate biases, RQ2 and RQ3 utilized the different Facebook dataset through random sampling on Facebook posts composed of 135 posts, including 45 posts from DPP, KMT and environmental group pages, which would keep the same article number as the three parameters to measure slant.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i Un. v. DOI:10.6814/NCCU202000381.

(5) The results showed that the Shenao Power Plant-related topics mentioned by all communicators (i.e., DPP/KMT political actors and environmental groups on Facebook, traditional news media and digital native news media) focused on three dimensions: Shenao power plant, politics and Government, and environmental concerns. The terms of two political parties (KMT and DPP) dominated the top-ten salient issues among all the communicators except for the environmental groups. The environmental groups on Facebook focused more on the environmental concerns in their top-ten salient issues. As for the correlation analysis, both traditional media and digital native media were significantly correlated to a greater degree to the environment network, rather than correlated to DPP or KMT actors. That is, what the news media reported on Shenao power plant issue was closely similar to what environmental groups discussed about, rather than political actors’ discussion focus. Both traditional news media and digital native media showed more partisan slant towards KMT actors. As for the partisan slant of news media brands, eight media leaned towards KMT actors (pro-KMT), but one digital native media and one traditional news media showed more correlated to DPP actors (pro-DPP). Regarding the result of RQ3, the pro-DPP and pro-KMT media shared similar crucial concepts: Shenao power plant, two major political parties (DPP and KMT), and the words related to the government. That is, traditional and digital native news media focused on Shenao power plant issue, partisan, politics, and environmental concerns, indicating that despite partisan leaning, both media types mainly focused on similar themes in two political parties. As for contributions, this study made efforts to fill the research gaps by investigating media slant, framing, semantic network analysis, Taiwan’s environmental concerns through semantic network analysis. As the measurements of media slant were still diverged (D'Alessio & Allen, 2000; Lichter, 2017), this research proposed to use the social media contents as parameters to measure the slant, extending the concepts of “partisan slant” or “media slant”. Additionally, this research extended the understanding of intermedia influences across media platforms through semantic way, providing more possibilities of using intermedia contents to measure partisan slant. Empricially, it explored the different framing and salient issues between news media social media, providing text-based research insights on Taiwan news research on referendum issues. With a two-system party politics, Taiwan encountered conflicts between the two major political parties (i.e., KMT and DPP), and based on the analysis of the Shenao plant case, the environmental issues turned into political framing in some reports on local news media.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i Un. v. keywords: partisan slant, semantic network analysis, social media, referendum, Taiwan. 2. DOI:10.6814/NCCU202000381.

(6) TABLE OF CONTENTS CHAPTER I - INTRODUCTION. 1. 1.1. Research Background. 1. 1.2. Research Purposes. 5. 1.3. Research Significance. 8. CHAPTER II - LITERATURE REVIEW. 10. 2.1. News Framing in Agenda Setting. 10. 2.2. Digital-Native Media and Online Traditional Media. 11. 治政大 2.4. Social Media Contents as Bias Parameter 立 2.5. Semantic Network Analysis in Framing and Agenda Setting. 13. 2.3. Media Slant and Bias. 17 19. ‧ 國. 學. 2.6. Shenao’s Project and Politics in Environmental News Frames. n. al. 3.2.1. Data Analysis Procedure. 31. er. io. 3.2. Data Analysis. 29. sit. 3.1.1. News Media Articles 3.1.2. Facebook Posts. 28. y. Nat. 3.1. Data Collection. 27. ‧. CHAPTER III - METHOD. 21. Ch. engchi. i Un. v. 34 35. 3.2.2. Unit of Analysis. 36. 3.2.3. Data cleaning before analysis: Frequency of term and co-word matrix. 37. 3.2.4. Semantic Network and Salient Concept Analysis. 40. 3.2.5. Correlation of Network Analysis. 42. CHAPTER IV - RESULTS. 45. 4.1. Descriptive Results: Trends of Shenao’s Issue. 45. 4.2. Semantic Networks Analysis: Facebook Actors and News Media. 49. 4.3. Partisan Slant of News Media: Network Correlation Analysis Results. 58. DOI:10.6814/NCCU202000381.

(7) 4.4. Semantic Network Analysis: Pro-DPP/KMT News Media in Shenao’s Issue CHAPTER V - CONCLUSIONS AND DISCUSSIONS. 62 67. 5.1. Conclusions. 67. 5.2. Discussions. 70. 5.2.1. The Descriptive Results and the Intermedia Contents. 70. 5.2.2. The QAP Correlation and Partisan Slant Analysis. 72. 5.2.3. The Semantic Network Analysis: Salient Concepts and Slant Correlation. 76. 5.3. Research Contributions. 78. 5.4. Limitations and Future Research. 81. References. 立. Appendix 1 - Boolean Keywords. 政治大. 85 96. ‧ 國. 學. Appendix 2 - Lists of DPP, KMT, and Environmental Groups Facebook Fan Pages. 98. Nat io. sit. y. 105. n. al. er. Networks. ‧. Appendix 3 - List of Top-100 Words of Greatest Document Frequency as Nodes of Semantic. Ch. engchi. i Un. v. 2. DOI:10.6814/NCCU202000381.

(8) LIST OF TABLE AND FIGURE Tables Table 1. Events of Shenao Power Plant Issue. 26. Table 2. Selected Online News Media by uMiner. 30. Table 3. Counts of Selected Facebook Posts and Fan Pages by QSearch. 32. Table 4. Sorts of Selected Partisan Facebook Fan Pages by QSearch. 33. Table 5. Counts of Selected Facebook Posts and Fan Pages through Random Sampling for Measuring Partisan Slant as the Parameters. 34. 政治大. Table 6. Top-Ten Salient Concepts with Greatest Eigenvector Centralities: Facebook Actors. 立. and News Media. 49 58. Table 8. List of Pro-DPP and Pro-KMT News Media. 63. ‧ 國. 學. Table 7. QAP Correlation Analysis between News Media Networks and Parameter Network. sit er. al. n. Figure 1. Research Design. io. Figures. 64. y. Nat. KMT News Media. ‧. Table 9. Top-Ten Salient Concepts with Greatest Eigenvector Centralities: Pro-KPP and Pro-. Ch. engchi. i Un. v. 28. Figure 2. Word Segmentation Processed by JiebaR. 38. Figure 3. Co-word Matrix Processed by R Studio. 40. Figure 4. Time Series of Facebook Posts about DPP, KMT and Environmental Group. 46. Figure 5. Time Series of News Media Posts on Traditional and Digital Native News Meida 47 Figure 6. Time Series of Intermedia Posts on Facebook and News media. 48. Figure 7. The Semantic Networks of DPP, KMT, and Environmental Groups Actors. 54. Figure 8. The Semantic Networks of Traditional News Media and Digital Native News Media 57 Figure 9. The Semantic Networks of Pro-DPP and Pro-KMT News Media in Shenao issue 66. DOI:10.6814/NCCU202000381.

(9) CHAPTER I INTRODUCTION. 1.1. Research Background As news is a representation of the world, it involves selections for any representation, which would cause bias whenever news media decide what to present on coverage (Schudson, 2003). As previous studies suggested, media bias was considered “more or less” left or right leaning. 治政 of “center” (Gentzkow, Shapiro & Stone, 2015). Rather than大 the “honesty or accuracy”, media 立. ‧ 國. 學. bias or slant is more like a “taste or preference” as specific issues or its actors are described in. news outlets (Groseclose & Milyo, 2005). News slant or bias exists in nowadays society,. ‧. although news media are expected to release objective reporting. Partisan slant was believed. y. Nat. io. sit. to exist in some mainstream media which are in favor of one political party (Merkley, 2019).. n. al. er. Many scholars considered it difficult in operationalizing concepts of “bias” or “balance”. Ch. i Un. v. and measuring them precisely, as media slant might be triggered by political concerns or media. engchi. routines (Hopmann, Van Aelst & Legnante, 2012; Lichter, 2017). In previous studies, some scholars proposed several ways to measure media slant, exploring factors that might shape news stories, such as media’s mediatization, commercial needs, economic marketplace, governmental pressures, public opinions, and so on (Gentzkow et al., 2015; Gentzkow & Shapiro, 2010; Groseclose & Milyo, 2005; Entman, 2007; Gerth & Siegert, 2012; D'Alessio & Allen, 2000; Lichter, 2017). Several researchers focused on content-based analysis, considering. 1. DOI:10.6814/NCCU202000381.

(10) that the framing or the text-based factors of contents are crucial to media slant. This study would adopt content-based and framing analysis as be the main focus in this study. In a democratic society, politics in Taiwan is considered bi-polarized by major political parties of Kuo Ming Tang (KMT) and Democracy Progress Party (DPP), which consists of the “pan blue (KMT) leaning” and the “pan green (DPP) leaning” camps (Hsiao, 2014). Different from American politics with “liberal-conservative” or “left-right” political ideologies, Taiwanese two-party politics system is primarily stemmed from the party’s stance of crossstrait (“independence-unification”) relationship with China (Hsiao, 2014). While KMT is. 治政 generally considered pro-reunification with China, DPP is 大 in support of pro-independence of 立. ‧ 國. 學. Taiwan. Under this circumstances, issues related to political activities such as elections or referendums, are easily framed by media as strategic game in news coverage, in which related. ‧. to winning and losing, political actors, or campaign strategies, rather than substance of specific. y. Nat. io. sit. issues (Aalberg, Strömbäck & De Vreese, 2012). That is, news media might be highly relevant. n. al. er. to political concerns instead of issue foci, or even leaning on one-sided ideology.. Ch. i Un. v. Some scholars (Bolsen & Matthew, 2018) probed into what and how news media selected. engchi. contents as salient issues. As for how media portrayed the issues, some scholars suggested that it is a crucial concern of media’s news selection on partisan conflicts, since they may have better news worthy than focusing only on issue consensus (Baum & Groeling, 2008; Iyengar & Hahn, 2009). In Robison and Mullinix's study (2016), news slant or polarization in news framing was stemmed from elite’s ideological commitment and their strategic incentives from whose political stance. Brants and Voltmer (2011) suggested the interrelationship of political actors and media gave rise to the strategic news for political marketing and strengthen. 2. DOI:10.6814/NCCU202000381.

(11) “candidate-centered, image-driven, polarized and spectacular” news, but “less organized around issues and ideologies” (p.5). This study would conduct a specific investigation on identifying “what” news media chose to expose in their frames of an environmental and energy policy issue at 2018 referendum, and measuring partisan tendency. In Taiwan, the shortage of electricity is an urgent but long existing problem. Shenao power plant is a thermal power station located in northern Taiwan. Early in 2015, Taiwan’s Chinese National Federation of Industries (CNFI) identified “five shortages” of Taiwan, referring to the lack of land, water, power, manpower, and talent,. 治政大 from over-seas declined which led the deteriorating economic competition as investments 立. ‧ 國. 學. (Chinese National Federation of Industries, 2018; Ferry, 2018a). Parallel to two-system politics circumstances, environmental issues are long tied up with citizens’ political ideologies. Since. ‧. Taiwan’s struggle on electricity shortage was under vigorous debates, the Shenao power plant,. y. Nat. io. sit. which the petition was signed for referendum on November 24 2018, generated more. n. al. er. attractions and discussions (Ferry, 2018b).. Ch. i Un. v. In the end of June in 2018, Environmental Protection Administration (EPA) officially. engchi. announced that the project of Shenao power plant expansion was approved by environmental impact assessment review. On July, several environmental NGOs, (including Anti-Shenao Plant Self-Help Group, Environmental Rights Foundation, Wild Heart Taiwan, Greenpeace, and so on) initiated a protest nearby Executive Yuan, appealing to withdraw the Shenao power plant project for air equality and ecological protections (Wild Heart Taiwan, 2018; Lin, 2018a; Chang & Teng, 2018). On October 12, a month before the referendum (November 24), Taiwan premier announced that the government was calling off the Shenao power plant expansion. 3. DOI:10.6814/NCCU202000381.

(12) project (Teng, 2018). The scrapping of Shenao power plant project was seen as the consideration of the votes of election in 24 November (Apple Daily, 2018). The election was considered a prelude for the presidential election in 2020, in which the ruling party DPP was facing tough challenge from the opposition KMT (Aspinwall, 2018). News outlets of the Shenao power plant showed political struggles and conflicts between KMT and DPP, despite being an environmental issue. The project was interpreted as a strategy between the politics game between KMT and DPP in several local news (ETtoday News, 2018; Hsu, 2018; Chang, 2018). This research considered the Environmental Protect Administran’s. 治政大Shenao power plant expansion (EPA) official announcement as a crucial starting point, that 立. ‧ 國. 學. project approved by environmental impact assessment review on June 28. It would be the starting point for the time frame of the analysis in this study. From this time, the government. ‧. officially announced the agreement of Shenao power plant project. Besides, the time frame of. y. Nat. io. sit. analysis was set to end by November 30, 6 days after the referendum occurred.. n. al. er. Besides the partisan conflicts of Shenao power plant issue suggested in news outlets,. Ch. i Un. v. news frame of political conflict or strategy in environmental issues was analyzed and discussed. engchi. in previous studies (Bolsen & Shapiro, 2018; Nisbet, 2010; O’Neill, Williams, Kurz, Wiersma & Boykoff, 2015). News frames of political conflict or strategy in environmental issues (such as climate change, or international environmental summits) highlighted the conflicts among elites or political actors, such as winners or losers in political strategies, or battle of groups as games behind policies (Nisbet, 2010; Bolsen & Shapiro, 2018). Along with the previous studies and local news outlets, Shenao power plant project as an environmental issue tied with partisan discourses among news media would be designed as the case in this study.. 4. DOI:10.6814/NCCU202000381.

(13) This research would not only analyze the news media contents, but also the posts from political actors and environmental groups on Facebook. The comparison and observations on frames and salient concepts among different communicators, between news media, political actors, environmental groups and news media would be the first focus in this study. As for the the partisan slant of news media, the second focus in this research, semantic networks made of KMT/DPP political actors’s Facebook posts would be utilized as the parameter to measure the partisan slant in different types of news media. In addition, the semantic network made of Facebook posts from environmental groups would be used as another parameter. Although. 治政大 and news media was not the measuring the network correlation between environmental groups 立. ‧ 國. 學. main research purpose in this research, adding the environmental semantic network as another parameter would help to observe how the discourse of news media correlated to the partisan. ‧. discourse. The network correlation analysis would explore the partisan slant and the correlation. y. Nat. n. al. er. io. sit. to environmental discourse in different types of news media and each news media brand.. 1.2. Research Purposes. Ch. engchi. i Un. v. Semantic network analysis would be used in this study. As a text-based analysis, semantic network analysis measures the relationship of the words, focusing on the shared meaning and the relationship of concept constructions in word co-occurrence pairs as they form in groups or clusters (Rice & Danowski, 1993; Doerfel & Barnett, 1999). It would map the salient issues in different types of news media (traditional and digital native) and social media contents. This research section would depict the dominated issues framed in news coverage. It would help to. 5. DOI:10.6814/NCCU202000381.

(14) examine whether politics concern or issue based message would dominate news coverage, as the Shenao power plant issue as a referendum issue related to political activities. Next, it would explore the existing of partisan slant in news contents on the issue, from the aspect of media’s content selections that caused selection bias. From the past studies that measured the partisan slant through content-based way, for example, Gentzkow and Shapiro (2010) constructed a new index and compared the phase frequency of newspaper with Republican and Democrat congressional documents, to identify the partisan slant of news media. However, these studies measuring partisan slant only based on frequency of mentions,. 治政大 of agenda setting or framing, in order to cover large amount of news contents. From the aspect 立. ‧ 國. 學. it could compensate for staying at level of frequency. The aspect of “agenda bias”, which. measured slant based on agenda level, covering the level of news-story understanding about. ‧. how journalists or news media selected the issues or concepts (Eberl, Boomgaarden & Wagner,. y. Nat. io. sit. 2017). However, many scholars used manual content analysis to measure partisan slant from. n. al. er. aspect of agenda setting or framing, in order to understand the news story in news outlets.. Ch. i Un. v. To fill the research gap of going beyond the frequency level and simultaneously. engchi. understanding selected issues regarding news framing when measuring the partisan slant, this research proposed to use semantic network analysis to identify the salient issues, and measure the partisan slant of news media. Partly based on Gentzkow and Shapiro’s research (2010), this research would also consider contents from two political parties (KMT and DPP) as the partisan parameter, but it would take social media contents to measure the partisan slant. From the aspects of scholars, social media gradually reshape the formation of journalism, as the reciprocal information flow between social media and news media was observed by scholars. 6. DOI:10.6814/NCCU202000381.

(15) (Neuman, Guggenheim, Jang & Bae, 2014; Jang, Mckeever & Kim, 2019; Wang & Guo, 2018). As mentioned in the previous section, the semantic network made of Facebook posts from environmental groups would be used as another parameter, which would help to observe how the discourse of news media correlated to environmental discourse directly, and the partisan discourse indirectly. In order to investigate news contents showing political struggles in this referendum issue, it would be the first focus (RQ1) to explore the salient issue and framing in different types of media (online news media and social media) regarding the Shenao power plant. For the online. 治政 news media, it would investigate how two different types of大 news media (traditional and digital 立. ‧ 國. 學. native) portrayed the issue. In addition, this study would consider social media content a crucial role in this study. Due to the ubiquity of social media use, people gradually tend to obtain their. ‧. news information on social media (Albalawi & Sixsmith, 2015). It would also be a research. y. Nat. io. sit. focus on agenda setting of political actors (politicians or political organizations) and. n. al. er. environmental groups in this issue, through analyzing salient concepts of posts that mentioned. Ch. i Un. v. Shenao power plant project from KMT/DPP political actors and environmental groups. That. engchi. is, this study would first (RQ1) compare salient issues portrayed not only in the two types of online news media (traditional/digital native), but also the Facebook contents in two-party fan pages, along with environmental groups fan pages. The second research question (RQ2) would compare the partisan slant of two types of online news media (traditional/digital native). In first part of second research question (RQ2a), it would measure the partisan slant of the two types of news media. In the second part (RQ2b), it would identify each partisan slant of all news media whether belonging to traditional or. 7. DOI:10.6814/NCCU202000381.

(16) digital native news media. The social media contents refer to the DPP/ KMT political actors’ official fan page on Facebook, the same dataset from the first research part (RQ1). The semantic network correlation analysis (Quadratic Assignment Procedure, QAP) (Krackhardt, 1988) would define the partisan leaning on online news media, by calculating the correlations of contents between Facebook posts and news media. In both RQ2a and RQ2b, this research would include the semantic network of environmental groups as another parameter, for the nonpartisan comparison as mentioned before. The third part of the research (RQ3) would investigate the salient issue and framing upon the news media labelled partisan leaning. Similar. 治政大 would be used to identify the to the first research process (RQ1), semantic network analysis 立. ‧ 國. 學. salient issues of news media. Different from the RQ1, two groups of news media would be categorized upon the partisan leaning that would be process in RQ2. On top of this, it can. ‧. compare the framing and salient concept under the classification of types of media. y. Nat. n. al. er. io. sit. (traditional/digital native) and partisan leaning (pro-KMT/pro-DPP).. 1.3. Research Significance. Ch. engchi. i Un. v. As for the research significance, first of all, this research explored the applications of semantic network analysis, partisan slant research, and framing analysis. Regarding semantic network research, the applications of the semantic correlation analysis (QAP) on partisan slant research was a research focus. It would provide another way to measure media slant, upon the text-based level, and help this research to explore the possibility of doing quantitative content analysis. Second, this study would explore Taiwan news research on referendum issues, based on the investigation on Shenao power plant issue. Due to the two-party system and politics. 8. DOI:10.6814/NCCU202000381.

(17) struggle in Taiwan, news of referendum issues was easily laden with conflicts of two parties rather than issue-based focus. It would be another focus to analyze Taiwan referendum news research through semantic scopes. The application of intermedia contents would be the third focus in this study. For one thing, through using the semantic way, this research would investigate more possibilities of using intermedia contents to measure partisan slant. For another, this research would explore the different framing and salient issues between news media and social media upon the text level.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat al. Ch. engchi. i Un. v. 9. DOI:10.6814/NCCU202000381.

(18) CHAPTER II LITERATURE REVIEW. 2.1. News Framing in Agenda Setting Framing defined by Entman (1993) is “to select some aspects of a perceived reality and make them more salient in a communicating text”, referring as “selection” and “salience” (p. 52). It provides the scheme about the issue that media put emphasis on. Framing allows to it locate, perceive, identify, label occurrences in terms, providing the understanding and shape of an. 治政 issue, considered schemata, which give the interpretation of大 specified cultural central elements 立. ‧ 國. 學. (Goffman, 1974). To frame is to select what is perceived reality and make them more salient in. communication, to “promote a particular problem definition, causal interpretation, moral. ‧. evaluation, and/or treatment recommendation” (Entman, 1993, p.52). It is an “unavoidable. y. Nat. io. sit. reality of the public communication process”, as journalists, experts, or advocates choose and. n. al. er. conceptualize what to raise the awareness or mobilize the public (Nisbet, 2010, p.44).. Ch. i Un. v. Framing was seen as an extension of agenda-setting, for that it involves the attribution. engchi. about making an issue salient through selections (Scheufele, 2000; Chyi & McCombs, 2004). As Chyi and McCombs (2004) suggested, the object salience was defined as “a public issue, public figure, or any other topic” (p. 22) to be the first level of agenda setting, referred to the important topic or issue of media agenda. The second level of agenda setting involved attributions that how media selected and presented the salient objects, and how these selections influenced individual’s perceptions, such as framing and priming (Chyi and McCombs, 2004).. 10. DOI:10.6814/NCCU202000381.

(19) The framework could suggest various positions, rather than one common perspective, from different aspects interpreted by media or experts and how controversy of the issue is (Gamson & Modigliani, 1989). It could provide the public the shape of an issue, and help people as a tool to draw communication strategies and resolutions of problems (Van Gorp, 2010). As Van Gorp (2010) suggested, through symbolic stereotypes, values, archetypes, myths or narratives, it would help to understand how journalism selected notions, interpreted the event, and framed news sources and issues. As framing was referred as selection and salience,. 治政大 moral judgement, and suggest for which journalists define problems, interpret causes, make 立. ‧ 國. 學. remedies (Entman, 1993), to analyze the framing not only focus on the objects journalists. emphasize on, but look into the relationship of the messages at an social level of how these. ‧. selections constructed. This study would take Entman’s (1993) interpretation of framing as the. y. Nat. io. sit. core basis, that the framing referring as “salience” and media’s “selection”. Based on the theory,. n. al. er. this research would investigate framing of a referendum issue through the semantic network. Ch. i Un. v. analysis, which could investigate the salient issues based on frequency of word co-occurrence.. engchi. The word co-occurrence frequency would be used to measure the words degree of salience through several metrics, and would help to explore the “saliences” of the issue among media contents, and communicators’ “selections” of this issue.. 2.2. Digital-Native Media and Online Traditional Media Due to the ubiquity of Internet and mobile devices, many traditional news struggle to survive the decreasing revenue, adapting to new media environments (Nee, 2013), as audiences can. 11. DOI:10.6814/NCCU202000381.

(20) consume their news or first-hand information via social media sites or online (Wu, 2016). The traditional media make effort on adapting to the web environment (Nee, 2013). Also, the new environment media gave birth to “digital-native” or “digital-born” media. Wu (2016) defined the “digital natives” as “entirely born and grown online”, which might start as online content aggregators making curation or distributing viral contents (p. 131). Previous studies analyzed and made discussions on traditional legacy news media and digital native news media, in journalistic or commercial aspects (Cui & Liu, 2017; Wu, 2016; Painter, Kristiansen & Schäfer, 2018; Chen, Shi, Guo, Wang & Li, 2019), branding differences (Arrese & Kaufmann, 2016).. 立. 政治大. ‧ 國. 學. On the other hand, the digital-native news media considered by other scholars as nonprofit attempt to reshape the journalism through public serves reporting and audience. ‧. engagement (Nee, 2013). Harlow and Salaverría (2016) showed the competing relationship. y. Nat. io. sit. between traditional news media and online-native news websites by depicting how online-. n. al. er. native sites put efforts to renovate outdated patterns of journalism, engaging between. Ch. i Un. v. journalism, alternative media, and even activism. In Billard’s research (2019), the intermedia. engchi. research suggested causal effects of digital- native media on legacy press media in a specific issue. This research would compare the phenomenon of partisan slant between digital-native media and online traditional media. It would compare how the digital-native media and online traditional media suggested different salient issues on referendum issue, from a journalistic aspect. Under the political communication contexts, either scholars believed that the technology advances make digital native media decrease news equality compared to legacy. 12. DOI:10.6814/NCCU202000381.

(21) news because of the profit and growth needs, or renovating the journalism as they put effort to public concerns, the comparison between these two sort of media still worth investigations. The partisan bias exploration in textual level, words and terms formatting semantic networks would provide different landscape of comparison in the two types of media.. 2.3. Media Slant and Bias Slant or Bias is existed in several aspects in media, not only scholars focused generally on “media bias”, but also investigated in different scopes such as partisan bias (Merkley, 2019),. 治政大 2007), and so on. From the ideology bias (Jamieson & Cappella, 2008), framing bias (Entman, 立. ‧ 國. 學. past studies, media bias, slant, or partisan bias were in widely discussion (Bourgeois, Rappaz & Aberer, 2018), and it seemed not attributed to a major theorist (D'Alessio & Allen, 2000).. ‧. Scholars proposed their definitions and interpretations on this concept. Media bias is “nothing. y. Nat. io. sit. to do with the honesty or accuracy of the news outlet. Instead, our notion is more like a taste. n. al. er. or preference” as Groseclose and Milyo depicted (2005, p. 1204). In Entman’s study (2007), it. Ch. i Un. v. distinguished “bias” and “slant” differently, while “framing bias” was “consistent patterns in. engchi. the framing of mediated communication that promote the influence of one side in conflicts over the use of government power” (p. 166), the “slant” was “the framing favors one side over the other in a current or potential dispute.” (p. 165) (Entman, 2007). On the basis of Entman’s definition of “bias” and “slant”, this study would consider “slant” as the major concept, since it is more matched with the context of the study. Since 1970s to 1990s, slant or bias was widely investigated in newspaper and TV. In recent years, the core definition, measurement or impact of media slant are still diverged. 13. DOI:10.6814/NCCU202000381.

(22) (Lichter, 2017). As several factors are considered to cause partisan media, to operationalize media bias accurately is not that easy. Several factors were believed by scholars that could affect media slant, such as media’s mediatization, commercial needs, economic marketplace, governmental pressures, public opinions, etc (Gerth & Siegert, 2012; D'Alessio & Allen, 2000; Lichter, 2017; Groseclose & Mily, 2005). Hopmann et al. (2012) believed that media balance was defined by “political system” and “media routines”. The former involved how the contents cover and treat political actors equally that often occurs in two-party system, and the latter refers to media’s routine news selection based on news values or newsworthiness.. 治政大or bias in aspects of economics, In recent years, some scholars investigated media slant 立. ‧ 國. 學. proposing economic bias models on the basis of media’s market demand and supply (Gentzkow. & Shapiro, 2010; Lichter, 2017; Gentzkow et al., 2015). Besides, media balance is also affected. ‧. by sponsors, political institutions, think tanks or specific interest groups (Ferguson, 2016;. y. Nat io. sit. Groseclose & Milyo, 2005).. n. al. er. As media slant was believed to affect citizen’s party or position preference (Eberl et al.,. Ch. i Un. v. 2017), the framing of different news media media could explain how media sources select their. engchi. information and spin an event to the audience (Ferguson, 2016). Therefore, many researchers focused on the content-based analysis, that the framing bias research is the most, or the effects of the news coverage analysis. Despite the fact that media slant or bias might be attributed to complex factors, from aspects of media contents three sorts bias were developed. Previous scholars (D'Alessio & Allen, 2000; Eberl et al., 2017) categorized three common sorts of media bias “visibility bias (coverage bias)”, “tonality bias (statement bias)”, and “agenda bias (gatekeeping or selectivity. 14. DOI:10.6814/NCCU202000381.

(23) bias)". In news outlets, “visibility bias” was treated as a benchmark measured by relative amount of coverage on some politicians or parties (Boomgaarden & Wagner, 2017). Studies partisan balance adopting analysis of visibility bias are believed to be the most, as it is considered straightforward to measure (Hopmann et al., 2012). “Tonality bias” is measured by valences on political parties or actors, evaluating how they are covered, and “agenda bias” investigated the extent that media highlight specific issues or information in connection with some political actors or parties (Eberl et al., 2017). Scholars who focused on the media contents conducted content analysis on intensity of. 治政 campaign and style (Qin, Strömberg & Wu, 2018), agenda大 setting and framing (Chen et al., 立. ‧ 國. 學. 2019; Gerth & Siegert, 2012), or valence of the statements (Rafail & McCarthy, 2018) to. measure the slant in news outlets. Chen et al. (2019) compared the haze-related contents on. ‧. types of media (traditional vs social media) through agenda network analysis, exploring the. y. Nat. io. sit. attributes of haze either reflected threat or efficacy in two types of media.. n. al. er. Gentzkow and Shapiro (2010) proposed a media index to measure political news. Ch. i Un. v. contents’ attitudes towards Republican or Democrat congress documents. They constructed a. engchi. new index and compare the phase frequency of newspaper with Republican and Democrat congressional documents, to identify the partisan slant of news media. Another example, Gans and Leigh (2012) also measure news media’s political positions through media mentions of public intellectuals. However, these studies measuring partisan slant only based on frequency of mentions. At this level, they investigated the partisan slant from “visibility bias” aspect, through relative amount of coverage on some politicians or parties (Boomgaarden & Wagner, 2017; Eberl et al., 2017).. 15. DOI:10.6814/NCCU202000381.

(24) From the aspect of agenda setting or framing, it could compensate for staying at level of frequency. The aspect of “agenda bias”, which measured slant based on agenda level, covering the level of news-story understanding about how journalists or news media selected the issues or concepts (Eberl et al., 2017). For example, Gans and Leigh (2012) coded the political stories on a 5-point scale and calculate the mean score to measure the left-right degree of news media. Eberl and his colleagues (2017) compared the policy issues portrayed in campaign communications and in media outlets through manual content analysis knowing the universe of news stories. However, many scholars used manual content analysis to measure partisan. 治政 slant from aspect of agenda setting or framing, in order to 大 understand the news story in news 立. ‧ 國. 學. outlets.. In order to fill up the research gap of going beyond the frequency level and. ‧. simultaneously understanding the selected issues regarding to framing from news media, this. y. Nat. io. sit. research proposed to use semantic network analysis to identify the salient issues, and measure. n. al. er. the partisan slant of news media. The semantic analysis based on words and terms mixed the. Ch. i Un. v. way of investigating media slant through aspects of “visibility bias”, and “agenda bias”. engchi. (Boomgaarden & Wagner, 2017). The semantic network analysis involved “visibility bias” as networks are constructed by the frequency of word-occurrence that makes connections; and it involved “agenda bias” as the words provided the salient concepts or issues that communicators put emphais on. Partly based on Gentzkow and Shapiro’s research (2010), this research would also consider contents from two political parties (KMT and DPP) as the partisan parameter, but it would take social media contents to measure the partisan slant. From the aspects of scholars, social media gradually reshapes the formation of journalism, as the “multi-flow”. 16. DOI:10.6814/NCCU202000381.

(25) communications between social media and news media was observed by scholars (Hilbert, Vásquez, Halpern, Valenzuela & Arriagada, 2017; Shi & Salmon, 2018).. 2.4. Social Media Contents as Bias Parameter In the contemporary media environments, individuals can receive information and news from more options than before. From online news and social media, people were exposed to abundant information, which enhance their political knowledge and (Gil de Zúñiga, Weeks & Ardèvol-Abreu, 2017). Due to the ubiquity of social media use, non-elite individuals could. 治政 obtain, share instant news information and generate their大 voices, where they are easier to 立. ‧. ‧ 國. 2015).. 學. exchange ideas and have more opportunities to engage in discussions (Albalawi & Sixsmith,. While news media coverage set the salience of selected issues (Chyi & McCombs, 2004). y. Nat. io. sit. as the first level agenda setting, the salience of debates on politics or referendum was regarded. n. al. er. as the second level agenda setting (Wirth et al., 2010). In recent years, scholars combined the. Ch. i Un. v. concepts of “framing” and “agenda setting” between news media and social media, and coined. engchi. the term “frame setting” (Wang & Guo, 2018) to define the newly intermedia agendas relationship, due to the dynamic information flows (Neuman, Guggenheim, Jang & Bae, 2014; Jang et al., 2019), which went beyond the traditional ways to define framing and agenda setting. The intermedia agenda setting suggested that the information flow interworks effects on multi-channel digital media (Billard, 2019) where media contents influenced or set agenda from each other (Wang & Guo, 2018). Scholars found the dynamic and mutual information flow between social media and news media where ideas and agendas might be led from each. 17. DOI:10.6814/NCCU202000381.

(26) other rather than one-way transmitting pattern (Wang & Guo, 2018; Neuman et al, 2014). For example, Wang and Guo (2018) found a reciprocal relationship between the online news and Twitter, while they covered the issue of genetically modified mosquitoes through trend frequency observations and granger causality. Furthermore, Jang et al. (2019) found the corelation of intermedia information flow from social media to online mainstream news media. These findings provided support the rationale of this study to measure partisan slant through cross-media contents and use social media contents as the parameter to explore partisan leaning of news media.. 治政 Second, given the tight connections on social media as大 public communication sphere, the 立. ‧ 國. 學. nature of social media communication flows, as some scholars mentioned, had gone beyond. the traditional “two-step flow;”and rather, the “multi-step flow” or “one-step flow” was. ‧. proposed (Hilbert et al., 2017; Shi & Salmon, 2018). That is, the influencers or social media. y. Nat. io. sit. users possess more power on setting their agenda on issues. The traditional “two-step flow”. n. al. er. indicated that the information would first reach the opinion leaders before flowing to the mass. Ch. i Un. v. audiences (Katz, 1957). As a result of technological advancements, the “two-step flow” was. engchi. modified to be “one-step flow” or “network-step flow” during the social conflicts, as communicators were easier to influence their peers (Bennett & Manheim, 2006). Due to the prevalent social media use nowadays, those who could generate social cues were likely to have great influences on agenda setting of news media (Xu, Sang, Blasiola & Park, 2014). The key opinion leaders on social media therefore have great power of influences on public debates or campaign initiatives (Shi & Salmon, 2018; Nisbet & Kotcher, 2009).. 18. DOI:10.6814/NCCU202000381.

(27) Third, from the aspect of journalism, research on social media (e.g. Twitter and Facebook) as news sources suggested the use of social media source on journalism increased (Von Nordheim, Boczek & Koppers, 2018). A study in Czech Republic suggested that the domestic affairs were most likely sourced in news articles (Hladík & Štětka, 2017). In Taiwan, the scenario on how social media contents influences news making and production was also under discussion (Lee, 2016; Huang, J. E., 2018). On top of the reason mentioned, that the dynamic information flow between social media and news media, the greater influences on social media influencers, and the use of social media. 治政大contents on social media as the contents as news sources in journalism, this research used 立. ‧ 國. 學. parameter. To measure the partisan slant of online news media, this research would make semantic correlation analysis on news media contents with social media contents. As. ‧. intermedia communication flow are intertwined, contents on social media, from key opinion. y. Nat. n. al. er. io. sit. leaders, such as politics or political organizations, might also shape the news framing directly.. iv. n C hin Framing andUAgenda 2.5. Semantic Network Analysis Setting engchi. Semantic network analysis was defined as “the use of network analytic techniques on paired associations based on shared meaning,” to analyze text-based individual perceptions or organizations (Doerfel, 1998, p. 16). The text-based analysis is to measure the relationship of the words, and to what extent these words connected in a co-occur pairs as they form in groups or clusters (Rice & Danowski, 1993). Semantic network is similar to social network analysis because of their common technique of network analysis (Doerfel & Barnett, 1999). Differed from the social network analysis investigating the individuals bonds from intertwined. 19. DOI:10.6814/NCCU202000381.

(28) connections (Scott, 1988), the semantic network analysis focuses on the shared meaning and the relationship of concept constructions (Doerfel & Barnett, 1999). That is, “concepts” were treated as actors (nodes) composing the network in semantic network analysis, rather than individuals or organizations. Semantic network analysis applied in agenda setting and framing could be traced back to 2000’s. Before the prevalence of social media use, these studies were mainly based on the relationship of words or on traditional content analysis on texts (Doerfel, 1998). These studies commonly focused on identifying key issues in politics debates (Doerfel & Connaughton,. 治政大 Murphy, 2001; David, Atun & 2009), policy or expert discourse (Murphy & Maynard, 2000; 立. ‧ 國. 學. La Viña, 2012; Stewart, Gil-Egui, Tian & Pileggi, 2006), or news framing (David, Atun & La. Viña 2010; Schultz, Kleinnijenhuis, Oegema, Utz & Atteveldt, 2012; Van Atteveldt,. ‧. Kleinnijenhuis & Ruigrok, 2008; Jang, 2013; David, Legara, Atun & Monterola, 2014).. y. Nat. io. sit. Generally, semantic network analysis has two main directions: mapping the relationship. n. al. er. between semantic units through defining networks, and the association of actors with concepts (Yang & González-Bailón, 2017).. Ch. engchi. i Un. v. According to Murphy and Maynard (2000), words of policy camp spokespersons were analyzed to identify the discourse network and associated patterns. Besides the issue frames, some scholars investigated how the issues were set rhetorically and its political effects with ideological positions (Murphy & Maynard, 2000). Studies on news coverage mostly drew framing comparison on specified issue. For example, Tian and Stewart (2005) compared how two online news (BBC and CNN) framed the SARS crisis, in which frequency of concepts and cluster themes were included in semantic analysis. For those studies mapping the framing. 20. DOI:10.6814/NCCU202000381.

(29) issues and relationship network, a wider broad of interpretations would be explored wider panorama on socialcultural aspect (Stewart et al., 2006). Along with the emerge of Web 2.0, social media was born as the newly type of media in a decade. Also, it made the existence of digital-native media on the basis of mutual engagement. Scholars of semantic analysis started their journey on intermedia contents (e.g. social media vs. online news media) (Billard, 2019; Yoon & Chung, 2018; Jiang, Barnett & Taylor, 2016; Yoon & Chung, 2018; Neuman et al., 2014) or social media messages (Liu, Lai & Xu, 2018). Neuman and his colleagues (2014) explored the framing of 29 political issues between. 治政大 relationship of the two media traditional and social media with big data, and proved the mutual 立. ‧ 國. 學. agenda setting.. Based up these previous studies, this research would investigate salient issues of. ‧. intermedia contents, in order to explore the framing, agenda setting in different type of media. y. Nat. io. sit. upon one specific issue. Further, as far as the current understanding from previous studies, little. n. al. er. studies mixed framing and semantic network analysis to investigate slant on news media. To. Ch. i Un. v. make this breakthrough, not only this study would utilize semantic way to sketch up the salient. engchi. concepts through network analysis upon the designated issue, but also further explore the partisan slant of news media.. 2.6. Shenao’s Project and Politics in Environmental News Frames The shortage of electricity is an urgent but long existing problem in Taiwan. On 15 August, 2017, a massive power blackout occurred, in which the five-hour long blackout affected 151 companies and caused the loss of 2.89 million US dollars (Yu, 2017). The giant power blackout. 21. DOI:10.6814/NCCU202000381.

(30) marks the electricity outage and the urgent scenario in Taiwan’s energy policy, as the president, Tsai Ing-wen, planned to phase out nuclear power by 2025, and initiate an energy transition plan to build new electricity-generating plants using renewable sources (“A Massive Blackout,” 2017). In 2018, Taiwan Power Company (Taipower) announced to expand and reopen the “Shenao thermal power plant”, to meet the sustainability and security of electricity supply, as in the transition period of energy plan for renewable sources (Ministry of Economics Affairs, 2018). Shenao thermal power plant, located in New Taipei City, had operated since 1960 and. 治政大 plant in South Asia (Ministry was decommissioned in 2007, which was once the biggest power 立. ‧ 國. 學. of Economics Affairs, 2011). The expansion project of Shenao power plant was opposed by environmental groups and local residents due to the concerns of air pollution and harm to. ‧. marine lives, despite the approval of environmental impact assessment review by Government. y. Nat. io. sit. (Liao, 2018; Greenpeace, 2018; DeAeth, 2018). In April 2018, the KMT proposed a. n. al. er. referendum petition to oppose the Shenao power plant (Central Election Commissions, 2018a;. Ch. i Un. v. Lin, 2018b). The question asked people whether they agree the energy policy of stoping any. engchi. new construction or expansion of any coal-fired power plant — including the Shenao Power Plant (Central Election Commissions, 2018b). The referendum of Shenao power plant issue is part of the multi-question referendum in Taiwan local election in 24 November, 2018, where voters decided their community-level leaders. Also, people voted for 10 referendum questions, involving issues of energy policy, same-sex marriage, gender equity education, name of usage in international events, etc (“The 10 Referendum Questions”, 2018).. 22. DOI:10.6814/NCCU202000381.

(31) In the end of June, Environmental Protection Administration (EPA) officially announced that the project of Shenao power plant expansion was approved by environmental impact assessment review. On July 9, several environmental NGOs, (including Anti-Shenao Power Plant Self-Help Group, Environmental Rights Foundation, Wild Heart Taiwan, Greenpeace, and so on) initiated a protest nearby Executive Yuan, appealing to withdraw the Shenao power plant project for air equality and ecological protections (Wild Heart Taiwan, 2018; Lin, 2018a; Chang & Teng, 2018). Also, New Taipei City Government also filed an administrative appeal to oppose the Shenao power plant expansion project few days later (Huang, A., 2018).. 治政 On October 12, a month before the referendum 大 (November 24), Taiwan premier 立. ‧ 國. 學. announced that the government was calling off the Shenao power plant expansion project, as. the third liquefied natural gas (LNG) terminal in Taoyuan would supply the original demand. ‧. of electricity (Teng, 2018). Four days ago, on October 8, LNG terminal in Taoyuan passed the. y. Nat. io. sit. environmental impact assessment, and the deputy director of the Environmental Protection. n. al. er. Administration resigned then (Teng, 2018). The scrapping of Shenao power plant project was. Ch. i Un. v. seen as the consideration of the votes of election in 24 November (Apple Daily, 2018). As the. engchi. election in 2018 was considered a prelude for the presidential election in 2020, the ruling party DPP (pro-independence of Taiwan) was expected to face tough challenges from the opposition KMT (pro-reunification with China) (Aspinwall, 2018). The project was interpreted as a strategy between the politics game between KMT and DPP in several local news (ETtoday News, 2018; Hsu, 2018; Chang, 2018). News outlets of Shenao power plant showed politics struggle and conflict between KMT and DPP, despite being an environmental issue. The project was interpreted as a strategy between the politics game between KMT and DPP in several local. 23. DOI:10.6814/NCCU202000381.

(32) news (ETtoday News, 2018; Hsu, 2018; Chang, 2018). Table 1 showed main events of Shenao power plant issue. This research considered the EPA’s official announcement on June 28, 2018, as a crucial starting point that Shenao power plant expansion project approved by environmental impact assessment review. As such, June 28 would be the starting point for the time frame of the analysis in this study. From this time, the government officially announced the agreement of Shenao power plant project. Besides, the time frame of analysis was set to end by November 30, 2018, 6 days after the referendum occurred. Besides the partisan conflicts of Shenao power plant issue suggested in news outlets,. 治政 news frame of political conflict or strategy in environmental大 issues was analyzed and discussed 立. ‧ 國. 學. in past studies (Bolsen & Shapiro, 2018; Nisbet, 2010; O’Neill, Williams, Kurz, Wiersma & Boykoff, 2015). News frames of political conflict or strategy in environmental issues (such as. ‧. climate change, or international environmental summits) highlighted the conflicts among elites. y. Nat. io. sit. or political actors, such as winners or losers in political strategies, or battles of groups as games. n. al. er. behind policies (Nisbet, 2010; Bolsen & Shapiro, 2018). As for Shenao power plant issue, as. Ch. i Un. v. an environmental issue, local news outlets also depicted political conflicts between political. engchi. parties, suggesting that project was interpreted as a politics game due to the policy change and the criticism between KMT and DPP (ETtoday News, 2018; Hsu, 2018; Chang, 2018). As Bolsen and Shaprio (2018) suggested, political actors, journalists and advocacy groups would make their effort to generate consensus, in order to aggregate strong public support for actions or policies. For this concern, as shown in news outlets upon Shenao’s issue, political conflicts would use as the strategy for stronger public consensus to support or boycott related actions.. 24. DOI:10.6814/NCCU202000381.

(33) Along with the previous studies and local news outlets, Shenao power plant project, as an environmental issue tied with partisan leaning among news media, was designed in this study. Combined with the research intention of media slant, this research chose the referendum as a case to observe news selections and slant on the basis of an environmental issue. This research would not only analyze the news media contents, but also the posts from political actors, including the politicians or political organizations of two political parties, and environmental groups speaking for Shenao power plant (e.g. Environmental Rights Foundation, Wild Heart Taiwan, Greenpeace) on Facebook. It would not include the “Anti-. 治政大 groups initiating the antiShenao Power Plant Self-Help Group”, one of the environmental 立. ‧ 國. 學. Shenao power plant, as the Facebook fan page was closed. In order to understand what was discussed in different media (social media and news media) and communication roles under. ‧. Shenao power plant issue, the comparison and observations on frames and salient concepts. y. Nat. io. sit. among news media, and political actors, environmental groups on social media becamethe first. n. al. er. focus in this study. Exploring the partisan slant of news media would be the second focus in. Ch. i Un. v. this research. The semantic networks made of KMT/DPP political actors’ Facebook posts. engchi. would be utilized as the parameter to measure the partisan slant in different types of news media. In addition, the semantic network made of Facebook posts from environmental groups would be used as another parameter. Although measuring the network correlation between environmental groups and news media was not the main research purpose in this research, adding the environmental semantic network as another parameter would help to observe how the discourse of news media correlated to the partisan discourse.. 25. DOI:10.6814/NCCU202000381.

(34) Table 1 Events of Shenao Power Plant Issue Date March 14, 2018. April 16, 2018. Event Analysis of the environmental impact of Shenao Power Plant approved by Government through committee votes Anti-Shenao power plant press conference held outside Legislative Yuan by anti-Shenao power plant local groups and DPP and NPP legislators KMT legislators started a referendum signature to oppose the. April 24, 2018. construction or expansion of any coal-fired power plant (Shenao power plant included). New Taipei City Government filed an administrative appeal to oppose the Shenao power plant expansion project. sit. y. Nat. The third liquefied natural gas (LNG) terminal in Taoyuan. io. passed the environmental impact assessment. n. al. er. October 8, 2018. withdrawing the Shenao power plant expansion project. ‧. July 18, 2018. Review Environmental NGOs initiated a protest and appealed for. 學. July 9, 2018. 立. Project Approved by Environmental Impact Assessment. ‧ 國. June 28, 2018. 政治大. EPA official announcement: Shenao Power Plant Expansion. October 12, 2018. Ch. i Un. v. Taiwan premier announced to call off Shenao power plant expansion project. engchi. October 24, 2018. EPA revoked the Environmental Impact Assessment Review. November 24, 2018. 2018 Taiwanese Referendum. *Note: EPA=Environmental Protection Administration; DPP= Democracy Progress Party; KMT=Kuo-Ming-Tang; NPP= New Power Party (Ministry of Economics Affairs, 2018; Liao, 2018; DeAeth, 2018; Lin, 2018b; Central Election Commissions, 2018a). 26. DOI:10.6814/NCCU202000381.

(35) CHAPTER III METHOD. This research conducted content analysis with computer-assisted semantic network analysis (SMA). SMA is an approach to visualize text analytics, and build up relationship term networks for the framing analysis. It could provide both quantitative and qualitative research sights as the quantitative metrics are basis of constructing text models, and would support qualitative analysis on salient issues in networks (Drieger, 2013). The research questions and research. 政治大. design (Figure 1) are as follows:. 立. ‧ 國. 學. RQ1: What are the salient concepts in semantic networks of different types of news media. ‧. (online traditional news media, digital-native news media) and Facebook posts from KMT. y. Nat. io. sit. and DPP political actors, and environmental groups on the Shenao power plant issue?. n. al. er. RQ2: What are the partisan leaning of (a) the two types of news media (traditional news. Ch. i Un. v. media and digital-native news media) (b) each news media brand, labelled as pro-DPP or pro-. engchi. KMT news media by means of semantically correlated with Facebook posts from political actors with DPP or KMT party stance? RQ3: What are the salient concepts in semantic networks of labelled pro-DPP or pro-KMT news media through semantic network correlation analysis with Facebook posts?. 27. DOI:10.6814/NCCU202000381.

(36) Figure 1. Research Design. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat al. 3.1. Data Collection. Ch. engchi. i Un. v. This research would conduct analysis with two datasets: uMiner’s online media news articles and QSearch’s Facebook posts. As for the analysis time frame, it treated the EPA’s official announcement as a crucial starting point, that Shenao power plant expansion project approved by environmental impact assessment review on June 28. From this time, the government officially announced the agreement of Shenao power plant project. Therefore, both datasets would start from July 1. The data frame was set to end by November 30, 6 days after the referendum from 1 July 2018 to 30 November 2018, focusing on Shenao referendum issue. All 28. DOI:10.6814/NCCU202000381.

(37) the online news articles were collected from uMiner, a platform providing omni-media monitoring and analysis service launched by uMax Data Technology, a Hong Kong company focusing on data mining. The second dataset was composed of social media contents, from Facebook fan pages of KMT and DPP political actors and environmental groups. All the Facebook posts were provided by QSearch, a Taiwanese company providing Facebook focused data monitoring and analysis service.. 3.1.1. News Media Articles. 治政 Online news media articles from uMiner platform involved大 two-step procedure. First, uMiner 立. ‧ 國. 學. provided all news articles searched by Boolean logic key term compositions related to energy issues in Taiwan as many as possible, including nuclear energy, thermal power, solar power,. ‧. hydropower, and so on (see appendix 1), in order to make sure collecting all energy-relevant. y. Nat. io. sit. articles. Second, uMiner searched online news articles by Boolean key words only related to. n. al. er. “Shenao power plant”, and collected all the news article mentioning this issue (also see appendix 1).. Ch. engchi. i Un. v. There were 7,875 articles from 94 news channels collected by uMiner. Form those news articles collected, the selection of news media in this study was mainly based on Reuters Institute Digital News Report (Reuters Institute, 2019), which marked lists of online media brand that were most frequently used weekly, and most trustworthy online media brands with higher trust scores. That is, the selected media were top frequently used or most trustworthy online news media.. 29. DOI:10.6814/NCCU202000381.

(38) From Reuter’s report (2019), this study selected six traditional online news media, including TVBS News online, ETTV Financial news online, Apple Daily online, Sanlih ETelevision online (SETN), United Daily News online (UDN), and Liberty Times online; three digital news media, including NOWNEWS, Storm Media, ETtoday, and Newtalk. This study excluded some news media brands listed by Reuter (2019) referring to web portal affiliated media (Yahoo! News), and news media with extremely small amount of articles, such as The Reporter (only two articles provided by uMiner). In addition, this study also took reference on a media list from another report, Online News Professionalism Evaluation: Preliminarily. 治政 Rating Daily News Media Facebook Fan Pages, provided by大 Media Watch (2018), a Taiwanese 立. ‧. ‧ 國. native news media.. 學. nonprofit media education organization. From this report, this study added Newtalk for digital. Totally 2,627 news media article were collected, from six top traditional media brands. y. Nat. io. sit. including 1,597 articles, and four top digital native news media brands containing 1,030 news. n. al. er. articles. Table 2 shows the list of selected news media and the article counts.. Ch. Table 2. engchi. i Un. v. Selected Online News Media by uMiner News Media. Code. Traditional Online News Media. Article Counts. TN. TVBS News online ETTV Financial news online (EBC Financial news) (東森財經新聞) Apple Daily online (壹蘋果日報). TVBS. 185. ETTV. 32. APPL. 143. 30. DOI:10.6814/NCCU202000381.

(39) SETN - Sanlih E-Television online (三立新聞網). SETN. 265. UDN - United Daily News online (聯合新聞網). UDN. 829. Liberty Times online (自由時報). LIB. 143. Totals of Traditional Online News Media (n=2,627). 1,597 (61%). Digital Native Online News Media. DN. NOWNEWS (今日新聞). NOW. 223. Storm Media (風傳媒). STOM. 233. ETtoday (東森新聞雲). ETT. 372. NEW. 202. Newtalk (新頭殼). 立. 政治大. Totals of Digital Native Online News Media (n=2,627). 2,627 (100%). ‧. ‧ 國. 學. Totals of News Media Articles. 1,030 (39%). y. Nat al. er. io. sit. 3.1.2. Facebook Posts. n. All the Facebook posts were provided by QSearch Trend panel. Posts that related to Shenao. Ch. engchi. i Un. v. power plant issue was collected by boolean keywords “(深澳電廠 or 深奧電廠 or 深奧藍莓 or 深澳藍莓 or 深澳燃煤 or 深奧燃煤) -seg: media” (see appendix 1) during 1 July 2018 to 30 November 2018. The panel would help to filter all news media Facebook fan pages. Before data cleaning, totally 3,453 articles in 570 Facebook fan pages were collected from QSearch Trend panel. To answer RQ1, this study would code collected Facebook fan pages as KMT/DPP political actors, or environmental groups. As for partisan fan pages, only official KMT or DPP political actors’ fan pages would be selected, which refer to fan pages of politician affiliated 31. DOI:10.6814/NCCU202000381.

(40) with KMT or DPP in 2018, and KMT/DPP’s political organizations. Politicians’ fan pages were identified as they were created and operated by the politicians or their affiliated work team members. KMT/DPP official pages would be identified as they affiliate with KMT/DPP’s official website. For example, they attached KMT/DPP official website link on pages’ “About” section, or the links of Facebook fan page site were listed on KMT/DPP official website. Unofficial fan clubs of KMT/DPP party or politicians operated by their fans were not included. Since DPP was the incumbent party, this study also coded the government agency pages (such as Ministry of Enomic Affairs, Bureau of Energy) as DPP official pages. As for environmental. 治政大including energy sustainability, group, fan pages related to environmental concerns were coded, 立. ‧ 國. 學. environmental pollution, environmental protection, and so on.. Totally, 817 posts and 181 fan pages were selected for the first part of analysis (RQ1).. ‧. Among the selections, 550 posts were from 121 coded KMT fan pages, 45 posts were from 23. y. Nat. io. sit. DPP fan pages, and 222 posts were from 37 environmental group pages (Table 3) (see Appendix. n. al. er. 2 for the list of selected fan pages). Table 4 showed the count of different sorts of political actors among DPP and KMT pages.. Ch. engchi. i Un. v. Table 3 Counts of Selected Facebook Posts and Fan Pages by QSearch Post Counts. Fan Page Counts. (%). (%). Democracy Progress Party (DPP). 45 (6.5%). 23 (12.7%). Kuo Ming Tang (KMT). 550 (67.3%). 121 (66.9%). Environmental Group. 222 (27.2%). 37 (20.4%). Totals. 817 (100%). 181 (100%). Political Party. 32. DOI:10.6814/NCCU202000381.

(41) Table 4 Sorts of Selected Partisan Facebook Fan Pages Provided by QSearch Democracy Progress. Kuo Ming Tang. Party (DPP). (KMT). Sorts of Political Actors Politicians. 18. 83. Political Organizations. 3. 38. Government Organizations (DPP). 2. Totals. 23. 121. 政治大 parameter, to measure the partisan slant of news media. Three semantic networks would be 立 To answer RQ2, KMT, DPP and environmental group fan pages would be used as. ‧ 國. 學. created based on posts from coded KMT, DPP and environmental group pages. As for the sampling, this study would keep the same article number in these three stacks of pages, to. ‧. eliminate biases. Although it did not cause the bias on the matrix size since each matrix should. y. Nat. er. io. sit. keep the same matrix size referring same number of nodes before conducting QAP analysis, the sampling size of posts might cause differences on the abundance of terms among three. al. n. iv n C communicators as the parameter. For thishconcern, i U would keep the same sampling e n g cthishresearch size of each Facebook communicator. From the coding results of three stacks of pages (Table 3), only 45 posts mentioned Shenao’s issue from DPP, which gave the sampling size in this analysis part. Despite the lack of posts from DPP pages, it could still generate sufficient words after segmentation process. Except for DPP posts, this study conducted random sampling from KMT and environmental group posts. The random sampling could prevent biases from the selection of some news event where the news outlets might frame some terms particularly.. 33. DOI:10.6814/NCCU202000381.

(42) After sampling procedure, as Table 5 suggested, totally there were 135 posts, where 45 posts from DPP, KMT and environmental group pages. For analysis, the three semantic networks were separately created as partisan parameter, representing DPP, KMT, and one nonpartisan parameter representing environmental groups. In RQ3, it would use the same arranged dataset as RQ2, where 45 posts from DPP, KMT and environmental group pages would be used.. Table 5. 治政大 Random Sampling for Counts of Selected Facebook Posts and Fan Pages through 立. 45 (33.3%). 18. 45 (33.3%). 27. 45 (33.3%). -. Totals. 135 (100%). n. Environmental Group. al. Ch. e n g c h- i. Political. Organizations. Organizations. (DPP). y. Politicians. Government. 3. 2. 18. -. -. -. -. -. sit. io. Kuo Ming Tang (KMT). (%). Number of. er. ‧ 國. (DPP). Nat. Democracy Progress Party. Number of. ‧. Political Party. Post Counts. 學. Measuring Partisan Slant as the Parameters. i Un. v. 3.2. Data Analysis In this small chapter, the first two sections would introduce the whole data analysis procedure, and unit of analysis for whole analysis process. Then, the next three sections would describe the details of data cleaning, semantic network and salient concept analysis, and semantic correlation analysis.. 34. DOI:10.6814/NCCU202000381.